The Chinese laboratory DeepSeek has introduced a new feature for its DeepSeek R1 model, which focuses on reasoning processes. Now, it can simultaneously browse web pages and perform logical computations. This update enables the model to access up-to-date information from the internet in real time, using its advanced logical framework to provide accurate answers.
Users can take advantage of this feature via the DeepSeek API or the web chat platform by activating the “Deep Thinking” mode.
DeepSeek is a new player in the AI market, known for its focus on reasoning models. The R1 series, which includes R1-Zero and R1, uses unconventional training methods such as reinforcement learning without supervised fine-tuning (R1-Zero) and a hybrid approach that combines reinforcement learning with supervised fine-tuning (R1). These models are open to the community under the MIT license, emphasizing transparency and collaboration with the public. The company has also released smaller distilled versions of R1 to make advanced reasoning capabilities available on less powerful hardware.
